基于改进卷积神经网络的机器人避障路径规划  

Robot obstacle avoidance path planning based on improved convolutional neural network

作　　者：王思越 宋骊颖 刘俊森 WANG Siyue;SONG Liying;LIU Junsen(School of Electric Engineering and Automation,Shenyang Ligong University,Shenyang 110168,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,辽宁沈阳110168

出　　处：《电子设计工程》2025年第1期27-30,35,共5页Electronic Design Engineering

摘　　要：由于机器人在运动过程中,无法及时更新机器人位置方向和坐标,使得目标不是全局最小点,导致机器人无法有效避障。为此,提出了基于改进卷积神经网络的机器人避障路径规划方法。使用基于改进卷积神经网络的双线性内插方法,计算目标点坐标。通过动态窗口法评估函数,计算扩展距离。构建极大值损失函数,通过递减学习,使类别内的特征聚合度和类别间差异性达到最大。通过机器人在环境中的运动来估计机器人位置,计算机器人的平移速度、角速度,并更新机器人位置方向和坐标。构建改进后排斥函数,计算神经元中心点到目标神经元中心点的距离,规划避障路径。实验结果表明,该方法能够避过全部障碍物,且规划的起始点和目标点之间距离与实际距离一致。Due to the inability of the robot to update its position direction and coordinates in a timely manner during its movement,the target is not the global minimum point,resulting in the robot being unable to effectively avoid obstacles.Therefore,a robot obstacle avoidance path planning method based on improved convolutional neural networks is proposed.Using a bilinear interpolation method based on improved convolutional neural network,calculate the coordinates of the target point.Evaluate the function using the dynamic window method and calculate the extension distance.Construct a maximum loss function,which maximizes the degree of feature aggregation within the category and the differences between categories through decreasing learning.Estimate the robot’s position through its movement in the environment,calculate the robot’s translation speed and angular velocity,and update the robot’s position direction and coordinates.Construct an improved rejection function,calculate the distance from the center point of the neuron to the center point of the target neuron,and plan the obstacle avoidance path.The experimental results show that this method can avoid all obstacles,and the distance between the planned starting point and target point is consistent with the actual distance.

关 键 词：改进卷积神经网络 机器人避障 路径规划 全局最小点 

分 类 号：TN96[电子电信—信号与信息处理]